[
https://issues.apache.org/jira/browse/OAK-7512?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16493754#comment-16493754
]
Julian Reschke edited comment on OAK-7512 at 5/30/18 11:50 AM:
---------------------------------------------------------------
https://issues.apache.org/jira/secure/attachment/12925566/OAK-7512.diff seems
to make the test pass reliably.
Downsides:
- this is because all operations in the test happen in the same session, thus
making checkin timestamps unique per session is sufficient
- it throttles checkins per session to the resolution of
{{System.currentTimeMillis()}} - 1000/s in the best case, worse on Windows in
general
A proper fix would require to get rid of timestamp comparisons altogether. That
would require however a new way to persist the relations between parent and
child versions. Maybe we could leverage node store version identifiers, but
we'd still have problems comparing them across versions created on different
cluster nodes.
It might be possible to improve the patch by only throttling when ancestors or
descendants are known to be under version-control, which might not be true in
general.
Feedback appreciated. [~mreutegg], [~mduerig], [~anchela]?
was (Author: reschke):
https://issues.apache.org/jira/secure/attachment/12925566/OAK-7512.diff seems
to make the test pass reliably.
Downsides:
- this is because all operations in the test happen in the same session, thus
making checkin timestamps unique per session is sufficient
- it throttles checkins per session to the resolution of
{{System.currentTimeMillis()}} - 1000/s in the best case, worse on Windows in
general
A proper fix would require to get rid of timestamp comparisons altogether. That
would require however a new way to persist the relations between parent and
child versions. Maybe we could leverage node store version identifiers, but
we'd still have problems comparing them across versions created on different
cluster nodes.
It might be possible to improve the patch by only throttling when the parent
(or ancestors) are known to be under version-control, which might not be true
in general.
Feedback appreciated. [~mreutegg], [~mduerig], [~anchela]?
> RestoreTest.testRestoreNameJcr2 occasionally failing
> ----------------------------------------------------
>
> Key: OAK-7512
> URL: https://issues.apache.org/jira/browse/OAK-7512
> Project: Jackrabbit Oak
> Issue Type: Task
> Components: jcr
> Affects Versions: 1.6.11, 1.9.2
> Reporter: Julian Reschke
> Assignee: Julian Reschke
> Priority: Minor
> Attachments: OAK-7512.diff
>
>
> This tests occasionally fails for me:
> {noformat}
> testRestoreNameJcr2(org.apache.jackrabbit.test.api.version.RestoreTest):
> restore must restore child node version 1.0. expected:<1.[0]> but was:<1.[1]>
> {noformat}
> I *suspect* it is timing related (because it seems to happen more frequently
> when the system is busy).
> Last seen on 1.6 branch.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
